Ted
@edwardbenson.bsky.social
Founder @ Steamship - AI agents for the workplace.
Trying to learn jazz piano.
Probably camping.
I believe in you.
📍 DC, SF, Taiwan
Trying to learn jazz piano.
Probably camping.
I believe in you.
📍 DC, SF, Taiwan
Totally. Also unless you’re using a very differently aligned LLM, the evaluator often suffers from the same judgement errors.
Eg preferring the same troupes or succumbing to the same reasoning errors
Eg preferring the same troupes or succumbing to the same reasoning errors
December 7, 2024 at 11:17 PM
Totally. Also unless you’re using a very differently aligned LLM, the evaluator often suffers from the same judgement errors.
Eg preferring the same troupes or succumbing to the same reasoning errors
Eg preferring the same troupes or succumbing to the same reasoning errors
I wonder how low-level you could take that.
Trying to blend OCR and LLM output is what prompted the tweet. Using each to smooth the flaws in the other.
Eg OCR breaks text stupidly and makes blurry-vision errors.
LLM fixes that.. but then goes overboard and hallucinates semantically.
Trying to blend OCR and LLM output is what prompted the tweet. Using each to smooth the flaws in the other.
Eg OCR breaks text stupidly and makes blurry-vision errors.
LLM fixes that.. but then goes overboard and hallucinates semantically.
November 24, 2024 at 4:48 PM
I wonder how low-level you could take that.
Trying to blend OCR and LLM output is what prompted the tweet. Using each to smooth the flaws in the other.
Eg OCR breaks text stupidly and makes blurry-vision errors.
LLM fixes that.. but then goes overboard and hallucinates semantically.
Trying to blend OCR and LLM output is what prompted the tweet. Using each to smooth the flaws in the other.
Eg OCR breaks text stupidly and makes blurry-vision errors.
LLM fixes that.. but then goes overboard and hallucinates semantically.
Wow this looks really slick
November 21, 2024 at 11:27 PM
Wow this looks really slick